Author: Eugene Nalimov
Date: 12:03:43 07/02/99
Go up one level in this thread
On July 02, 1999 at 13:15:43, José de Jesús García Ruvalcaba wrote: >On July 02, 1999 at 12:45:47, Peter Kappler wrote: > >>On July 02, 1999 at 12:35:21, eric guttenberg wrote: >> >>>Peter, >>> >>>It isn't really that Hiarcs is "pondering" since neither program can think on >>>its opponent's time but the effect is much the same. >>>As was discussed in an earlier thread, H7.32 does not clear its hash tables >>>after each moves so if it correctly "predicts" F5.32's move, H7.32 will >>>pick up where it left off in its analysis from the previous move, which >>>will result in a deeper search. So it is not a bug but rather a feature of >>>H7.32 that gives it a boost in one-computer games over F5.32. >>>This at least is the theory presented by Jim Walker. In any event, the >>>results speak for themselves. There is a difference between one-computer >>>game results and two computer game results as to these 2 programs, >>>whatever the reason. >>> >>>eric >> >> >>Why is Fritz clearing the hashtables between moves? And why only in >>engine-engine matches on the same machine? Almost sounds like a bug... >> >>--Pete > > Fritz always clears its hashtables between moves, not only in engine-engine >games. > Fritz is a root procesor and it makes no sense for it to keep the hashtables >for another search with a different root position. I beleive that best move that is stored in hash tables often is the best move with different root position and piece/square tables - after all, majority of positions that are considered by program is very strange, and usually have tactical refutation. So, there is a sense not to clear hash table even for root processor - just ignore score, but reuse best move. From my experience, main benefit from hash table (at least in the middlegame) comes from better moves ordering. Eugene
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.